Merge branch 'main' into 4534

Author: roypat

CHANGELOG.md

@@ -42,6 +42,9 @@ and this project adheres to
   `mem_size_mib` and `track_dirty_pages` being mandatory for all
   `PATCH /machine-config` requests. Now, they can be omitted which leaves these
   parts of the machine configuration unchanged.
+- [#5007](https://github.com/firecracker-microvm/firecracker/pull/5007): Fixed
+  watchdog softlockup warning on x86_64 guests when a vCPU is paused during GDB
+  debugging.
 
 ## [1.10.1]
 

Cargo.lock

@@ -22,7 +22,7 @@ vmm = { path = "../vmm" }
 vmm-sys-util = "0.12.1"
 
 [target.'cfg(target_arch = "aarch64")'.dependencies]
-clap-num = "1.0.2"
+clap-num = "1.2.0"
 
 [features]
 tracing = ["log-instrument", "fc_utils/tracing", "vmm/tracing"]

src/snapshot-editor/Cargo.toml

@@ -319,6 +319,16 @@ impl Vcpu {
                 // If the emulation requests a pause lets do this
                 #[cfg(feature = "gdb")]
                 Ok(VcpuEmulation::Paused) => {
+                    // Calling `KVM_KVMCLOCK_CTRL` to make sure the guest softlockup watchdog
+                    // does not panic on resume, see https://docs.kernel.org/virt/kvm/api.html .
+                    // We do not want to fail if the call is not successful, because depending
+                    // that may be acceptable depending on the workload.
+                    #[cfg(target_arch = "x86_64")]
+                    if let Err(err) = self.kvm_vcpu.fd.kvmclock_ctrl() {
+                        METRICS.vcpu.kvmclock_ctrl_fails.inc();
+                        warn!("KVM_KVMCLOCK_CTRL call failed {}", err);
+                    }
+
                     return StateMachine::next(Self::paused);
                 }
                 // Emulation errors lead to vCPU exit.

src/vmm/src/vstate/vcpu/mod.rs

@@ -95,7 +95,9 @@ def test_coverage(monkeypatch):
         if not branch:
             branch = utils.check_output("git rev-parse --abbrev-ref HEAD").stdout
 
-        codecov_cmd = f"codecov -f {lcov_file} -F {global_props.host_linux_version}-{global_props.instance}"
+        # -Z flag means "fail on error". There's supposed to be a more descriptive long form in
+        # --fail-on-error, but it doesnt work.
+        codecov_cmd = f"codecov -Z -f {lcov_file} -F {global_props.host_linux_version}-{global_props.instance}"
 
         if pr_number and pr_number != "false":
             codecov_cmd += f" -P {pr_number}"